From Tacit Expertise to Operational Logic: LLM-Mediated Knowledge Codification in Resource-Constrained Organizations

Engineering notes

Engineering notes will be added by the aipentium editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为大语言模型、生成式AI、ChatGPT相关技术、计算机视觉、深度学习等高价值论文补充中文说明。

Original abstract

Across many kinds of small organizations, including specialty retailers, local clinics, family businesses, regional service providers, and NGOs, the most valuable operational asset is often the tacit knowledge held by one or two domain experts. The IS knowledge-management literature has long emphasized the challenge of converting tacit knowledge into explicit and shareable forms, most notably in Nonaka’s work on organizational knowledge creation. In operational software development, however, this conversion often requires more than documentation. It requires a knowledge-engineering process that turns expert explanations into organized rules, reviewable logic, and implementable system behavior. Small organizations often lack the specialist roles needed for this process. What appears to be changing is not the human-to-human interview that begins knowledge elicitation, but the cost of the downstream steps that organize, formalize, and implement the resulting knowledge. A recent industry deployment in a small specialty retailer illustrates this shift. The retailer’s product space requires customers to navigate a sequence of interdependent configuration choices that has historically required a salesperson. This decision logic was learned through years of practice and was concentrated in the proprietor’s experience. Rather than engaging a separate knowledge engineer, a single developer conducted elicitation interviews with the proprietor directly, preserving a human-to-human interview that the proprietor would have recognized from a traditional consulting engagement. The developer then used a large language model as a cognitive aid for organizing and refining the resulting notes into candidate configuration rules for review, and used LLM-assisted coding to implement those rules into a deployed selection interface. The deployed system itself is conventional, relying on structured drill-down navigation and deterministic configuration logic. The more interesting phenomenon is the compressed pipeline that produced it, in which a single LLM-augmented developer was able to take on tasks traditionally divided among knowledge engineering, rule formalization, and software development roles. The deployment is anecdotal but suggests a broader research direction that is both timely and tractable. Several questions motivate the desire for AMCIS audience feedback. How should developer-LLM collaboration during knowledge organization and rule formalization be structured so that the codified operational logic remains faithful to the original expert and trustworthy for downstream users? What does this compressed pipeline mean for classical knowledge-management frameworks, particularly Nonaka’s SECI model of tacit-to-explicit knowledge conversion, when LLM-augmented developers mediate part of the externalization process between expert explanation and explicit system logic? And how should systems produced through this compressed pipeline be evaluated when the goal is transferring expert capability into operational software rather than maximizing benchmark accuracy?

Links and sources

• PDF from original source

Comments